Anssi
@anssir.bsky.social
This was something - not for the costs (although impressive) but for the fact that not letting LLM stop thinking and inserting 'Wait' (instead of e.g. 'Hmm') makes it better. Love it!
Published some notes on S1, a new reasoning model that was fine-tuned from Qwen 2.5 32B on just 1,000 examples and $6 of compute
Includes notes on running it with Ollama and exploring those 1,000 examples with Datasette Lite simonwillison.net/2025/Feb/5/s...
Includes notes on running it with Ollama and exploring those 1,000 examples with Datasette Lite simonwillison.net/2025/Feb/5/s...
S1: The $6 R1 Competitor?
Tim Kellogg shares his notes on a new paper, [s1: Simple test-time scaling](https://arxiv.org/abs/2501.19393), which describes an inference-scaling model fine-tuned on top of Qwen2.5-32B-Instruct for ...
simonwillison.net
February 6, 2025 at 5:55 AM
This was something - not for the costs (although impressive) but for the fact that not letting LLM stop thinking and inserting 'Wait' (instead of e.g. 'Hmm') makes it better. Love it!